Why Does a BMS Keep Failing Its Dielectric Test? It Is Where the TVS Sits

INTEK (www.intek.vip) is an authorized distributor of AMSEMI (Anhui Anmei Semiconductor) TVS devices; datasheets, samples and FAE review for the 5.0SMDJ standard and foldback -N families are available on request.

One-sentence conclusion: before blaming the insulation, draw one line on the schematic: where do the two test electrodes actually connect, and which devices have their two terminals straddling exactly those points. If both TVS terminals sit on the same side, the device is invisible to that test - both ends are lifted together and it stays off. The moment one terminal reaches across to the other side (enclosure, low-voltage ground, isolated ground), the test voltage drives it into avalanche and the meter reads the breakdown current of a 5 to 100 V diode, not the insulation resistance. The question was never how big the TVS should be, but where its reference point sits. Withstand is the job of the insulation system; a TVS is a transient energy valve, and using a protection device as an insulator guarantees a failed test.

flowchart TD
    A["Dielectric or IR test fails<br/>leakage over limit or resistance too low"] --> B["Step 1: mark the two test electrodes<br/>on the schematic"]
    B --> C{"Do the TVS terminals<br/>straddle those two points"}
    C -->|Both on the same side| D["Device is invisible to this test<br/>look for real insulation faults: creepage, cleanliness, potting, harness"]
    C -->|One terminal crosses over| E["Step 2: break that TVS branch and retest in place"]
    E --> F{"Does it pass now"}
    F -->|Still fails| D
    F -->|Passes| G["Confirmed: the branch was absorbing test voltage<br/>part is fine, position is wrong"]
    G --> H{"Can the process break it during test"}
    H -->|Yes| I["Break during test<br/>zero-ohm link, jumper, fixture bypass<br/>plus a documented restore check"]
    H -->|No| J["Change topology<br/>TVS for differential mode only, common mode to GDT plus Y caps and bonding"]
    J --> K["Re-check whether AC test fires the GDT<br/>AC and DC tests must be designed separately"]

1. Two tests that get called the same name

What the line calls a dielectric test is really two different measurements, and a TVS is sensitive to both:

• Dielectric withstand (HIPOT): apply a voltage far above the working level between two points (typical engineering values are 500 V DC, 1000 V DC or 1500 V AC for one minute, or a shorter production-line dwell), and watch the leakage current against a threshold. It answers whether the insulation can hold that voltage.
• Insulation resistance (IR): measure resistance between the same points with a 250 V, 500 V or 1000 V megohmmeter and compare against a tens-of-megohms requirement. It answers how good the insulation is.

A TVS perturbs them differently. The withstand test watches current, so a device pushed into avalanche blows through a milliamp threshold instantly. The IR test watches resistance, so once the TVS conducts the meter reads tens or hundreds of ohms and the board is rejected. A reading of a few hundred ohms is therefore almost never a failed insulator - it is a semiconductor sitting across the measurement loop.

The actual test voltage and pass threshold come from the applicable safety standard and the company specification. This article does not interpret any standard; it only addresses how a TVS interacts with the test.

2. The arithmetic: why the TVS is always dragged in

A TVS on a BMS has a VRWM somewhere between 5 V and 100 V. The top grade in the AMSEMI 5.0SMDJ and foldback -N families is 100 V (5000 W grade); the 3000 W and 8000 W grades stop at 90 V. The test starts at 500 V.

Take the most common 5.0SMDJ54CA-N (VRWM 54 V, VBR 60.0 to 66.3 V, maximum clamp Vc 69.6 V, all on the 10/1000us convention, datasheet 8060055):

Voltage across the two terminals	Relation to the 54 V grade	Device state	What the tester sees
54 V (normal working limit)	Equal to VRWM	Off, IR limit 5 uA (25 C maximum)	Normal, reading unaffected
Around 70 V	Above VBR maximum of 66.3 V	Avalanche, clamping near Vc	Some current, still within device capability
500 V DC	About 7.5 times VBR	Deep breakdown, current limited only by loop and tester impedance	Voltage will not rise, leakage over threshold
1000 V DC	About 15 times VBR	Same, with dissipation rising accordingly	Fails, and long dwell can damage the part
1500 V AC (peak about 2121 V)	About 32 times VBR	Breakdown and recovery every half cycle, about 3000 events in one minute	Fails, with the part repeatedly stressed

Three readings follow from that table, and all three are routine misdiagnoses on the line:

1. Seven times VBR is not a little leakage, it is full conduction. The tester sees an avalanching diode, not a megohm insulator. No power grade and no clamp level changes that - a bigger grade only changes how much energy the part survives, not whether it conducts.
2. AC and DC are different tests. The peak of 1500 V AC is 1500 x 1.414, about 2121 V, and it crosses zero every half cycle: at 50 Hz one minute is 3000 half cycles, so 3000 avalanche events. A DC test is one continuous breakdown; an AC test is a repeated hammering.
3. A one-minute dwell is outside the datasheet envelope altogether. Ipp is a 10/1000us single pulse, roughly a millisecond; one minute is about sixty thousand times longer. Thermally this is not a bigger surge, it is a different regime.

3. The only criterion: do the terminals straddle the electrodes

Do not reason from which functional block the TVS belongs to. Reason from electrical nodes. The same part, in two positions, is either invisible or an automatic failure:

TVS position	Terminals across the test electrodes	Effect on the withstand or IR test	What to do
Across B+ and B- (differential)	No, both ends lifted together	None, potential difference near zero, device stays off	Keep it, this is the correct use
Sampling line to board ground B-	No, board ground is also on the high side	None	Keep it
Sampling line or bus to enclosure (PE)	Yes	Driven into avalanche, milliamp leakage, fails	Change topology, or break it during test
Across the isolation barrier	Yes	Device is forced to hold the whole isolation voltage	Move it off the barrier; that job belongs to the isolator

The last row deserves its own sentence: the isolation barrier is designed to hold withstand voltage, a TVS is not. A TVS placed across the barrier asks a 5 to 100 V part to do work that belongs to the isolator and the creepage design. It will fail, and once it fails short it bypasses the entire barrier - the isolation is gone. For how the three isolation ratings (UL1577 test value, working voltage VIORM, transient surge) differ, see What Does 5 kVrms on an Isolator Mean; for which side of the barrier an interface TVS belongs on, see Does an Isolated Interface Still Need a TVS.

4. Four fixes, in order of preference

Fix one: change the topology (the real solution)

Take common-mode protection away from the TVS:

• TVS for differential mode only: across B+ and B-, or from a sampling line to board ground. Both terminals are inside the high-voltage domain, so it never enters the withstand loop.
• Common mode goes to GDT, Y capacitors and structural bonding: the correct exit for board-level common-mode surge is shielding, equipotential bonding and cabinet ground, not a TVS to the enclosure. A Y capacitor gives a high-frequency common-mode path without a DC path, so its effect on the withstand test is far smaller than a semiconductor (capacitance and rating must still be checked against the applicable safety requirements).
• Sampling ports follow sampling-line rules: series limiting resistor plus a low-voltage ESD or TVS part, as described in Should BMS Cell-Sampling Lines Have a TVS. Those parts have a VRWM of only a few volts, but if one is tied from a sampling line to the enclosure, 500 V destroys it just the same - position remains the first constraint, not grade.

Fix two: break the branch during the test (smallest change)

Leave a deliberately breakable test point in the TVS branch: a zero-ohm link, a two-pin jumper, a test pad, or a fixture-side relay bypass. Break it for the withstand test, restore it afterwards.

Two companion actions are mandatory, and skipping either creates a new problem:

• A restore check. Break, test, restore is the step most often missed, and an unrestored board ships with no protection while looking perfectly normal. Make the restore traceable with ICT, continuity test or a tick on the process card.
• Put the bypass on the fixture drawing. Plenty of failures are not design problems at all: the fixture uses a full-board copper post or a wide pressure bar and clamps the TVS branch into the loop. "Probes touch the test electrodes only" belongs on the drawing, not in someone's memory.

Fix three: split the test into segments (most common compromise)

Break the pack into segments and judge the TVS-bearing segment at a lower level:

1. Cell side to enclosure: low-voltage segment, 250 V or 500 V IR.
2. Output side (PACK terminals) to enclosure: the withstand segment.
3. Cell harness to enclosure: its own segment, clamped separately.
4. Communication port to enclosure: its own segment, where the isolator itself carries the withstand.

The cost is more stations and a slower takt; the benefit is no board change and no part change. On high-voltage platforms (32S, 48S) the segment test must also respect the potential difference between modules - see TVS Architecture for 32S and 48S High-Voltage BMS.

Fix four: specify a higher-voltage TVS (does not work, do not try)

This one needs an explicit refusal, because it is the most intuitive and most time-wasting option:

• The part does not exist. Standard TVS reverse standoff voltage tops out around 400 V, and the 5.0SMDJ and foldback -N families used on a BMS stop at 100 V. There is no 500 V or 1000 V grade.
• The logic fails anyway. Even if a 1000 V grade existed, on a 48 to 100 V BMS bus it would never conduct and its clamp would be useless - money spent on a part that does nothing while occupying an SMC footprint.

5. The other direction: the test is consuming the TVS

Almost nobody accounts for this, and it is the root of "the new board passes, the reworked board does not".

Datasheet Ipp and Vc are single-pulse 10/1000us figures; the design premise is absorbing one millisecond-scale energy event. A withstand test is a minute-long continuous breakdown, and in the AC case it is a repeated breakdown every half cycle. The consequences appear in this order:

1. Reverse leakage IR rises. The 5 uA limit in the 54 to 95 V grades of the AMSEMI 5.0SMDJ and -N families is a fresh-part, 25 C, maximum figure. After sustained breakdown it moves up. The visible effects are higher standby drain on continuously biased nodes, drift on sampling lines, and a pack that gets blamed for self-discharge - the diagnostic chain is in Does the TVS on a BMS Leak Enough to Drain the Battery.
2. Breakdown voltage VBR drifts. A shifted VBR breaks the voltage window you calculated, and the clamp moves with it.
3. Breakdown short. This is the dominant failure mode of a TVS - nine out of ten go short, not open - and when it happens at a cell-terminal position the loop closes through the cell and the TVS without passing the MOSFETs or the shunt, so the BMS neither sees it nor can switch it off. See Shorted TVS: Failure Position and Fuse Coordination.

Three rules follow for production: never validate a part by repeated voltage application; cap the number of retests on reworked units; and re-measure IR and VBR at room temperature after the test. The same logic applies to isolators - repeatedly proof-testing an isolation barrier is deliberate accelerated ageing, as explained in Do Isolators Wear Out.

6. GDT and MOV carry the same trap

Changing the topology can move you from one pit into another:

• A GDT is not a free pass. DC sparkover voltage ranges from tens of volts to over a thousand depending on the part; any type whose sparkover sits below the test voltage will fire at 500 V DC just like a TVS. Once it fires it is close to a short, the tester fails immediately, and in the worst case it stays in follow current. The only types that avoid this have a sparkover above the test voltage - and those are usually too slow to protect against the surge you bought them for, which is exactly why GDTs are awkward on a BMS.
• An MOV starts leaking at test voltage. Leakage climbs steeply with applied voltage, and the magnitude depends on the ratio of varistor voltage to test voltage (check the datasheet). At 500 V or 1000 V a varistor sitting near the edge of its conduction region reads hundreds of microamps to milliamps - enough to blow a production threshold. For how MOV and TVS divide the work, see Can a TVS Sit Directly on an AC Input and How Surge Protection Is Layered.
• AC and DC withstand must be designed separately. The peak of 1500 V AC is 2121 V; a common-mode branch that passes 500 V DC may be a completely different problem under AC.

7. In the field: false alarms from online insulation monitoring

Systems with online insulation monitoring (IMD) meet the same conflict in another form: IMD keeps reporting low insulation, yet the insulation checks out fine.

The mechanism is the same. IMD injects a signal between the bus and ground and derives insulation resistance from the divider ratio; any TVS from bus to ground puts its reverse leakage in parallel with the measurement and pulls the reading down. Three trend-based checks separate the two causes:

1. The reading worsens with temperature. TVS leakage roughly doubles per 10 C, while a genuine insulation defect (moisture, contamination, creepage tracking) is far less temperature sensitive. A reading that is worse at midday than at dawn, or worse after the cabinet heats up, points at the device.
2. It recovers when the branch is broken. Lift the bus-to-ground TVS and retest; if the reading returns to normal, the device was eating the measurement.
3. All clusters degrade together. A real defect concentrates in one cluster or one section (moisture, condensation, foreign material), while TVS leakage lifts the whole system at once.

8. Five steps to get it right once

1. Mark the devices that cross the withstand path on the schematic. Highlight every part whose terminals sit in different potential domains: TVS, MOV, GDT, Y capacitors, isolators. That drawing should be an input to fixture design.
2. Low voltage before high voltage. Run 250 V or 500 V IR first (is there an obvious short or moisture), then withstand (can it hold). Reverse the order and every failure costs two runs.
3. Probes touch the test electrodes only. Not the TVS branch, not a full-board pressure bar. Put it on the fixture drawing rather than relying on the operator.
4. Log leakage trends, not just values. The difference between first and second measurement carries more information than the absolute number: a board whose second reading is clearly higher is being damaged.
5. Make the restore check a line on the process card. Whatever was broken must be restored item by item, with a traceable record.

9. Common traps

• Treating a few hundred ohms as failed insulation. A conducting semiconductor in the loop produces exactly that reading; break the device before concluding anything.
• Trying to solve it with a bigger power grade. Within one voltage grade Vc does not change and only Ipp rises, while the question here is whether the device conducts at all.
• Applying AC conclusions to a DC test, or the reverse. 1500 V AC peaks at 2121 V and breaks down every half cycle.
• Letting the fixture clamp the TVS into the loop. The most frequent source of false failures, and fixing the fixture costs an order of magnitude less than changing the board.
• Forgetting to restore the break. An unrestored board looks perfect and ships without protection.
• Putting a TVS across the isolation barrier. Barrier withstand is designed, not delegated to a 5 to 100 V part - and a short there bypasses the whole barrier.

FAQ

Q1: How do I tell whether the TVS is the reason a BMS fails the dielectric test?

Two checks settle it. First, mark the actual test-electrode connections on the schematic (usually the high-voltage side or the cell harness against the enclosure or low-voltage ground) and find every device whose two terminals span those points. Second, break that branch (lift one part, remove a zero-ohm link, or bypass it at the fixture probe) and re-run the test in place. If it now passes, the TVS was absorbing the test voltage; if it still fails, the insulation really is at fault - check creepage, board cleanliness, potting and harness routing. The test voltage is in the 500 V DC to 1500 V AC range, far above any BMS TVS VRWM, so a straddling position guarantees avalanche. That is not a defective part, it is a topology error.

Q2: A bus TVS sits between B+ and B-. Does it affect a line-to-enclosure withstand test?

No. The only criterion is whether both terminals land on the two test electrodes. A bus TVS across B+ and B- is differential mode: during a common-mode test to the enclosure, B+ and B- are lifted together, the potential across the device stays near zero and it remains off, so it neither conducts nor disturbs the reading. The same applies to a TVS from a sampling line to board ground (😎 as long as board ground is not bonded to the enclosure. What fails is a common-mode protection position with one terminal on the high side and the other on the enclosure or isolated ground, plus any device across the isolation barrier.

Q3: Can I just pick a TVS rated for 1000 V and avoid the problem?

That path does not work. The part does not exist - standard TVS reverse standoff voltage tops out around 400 V, and the 5.0SMDJ and foldback -N families used on a BMS stop at 100 V with no 500 V or 1000 V grade. And the logic fails anyway: even a hypothetical 1000 V grade would never conduct on a 48 to 100 V bus, with a clamp far too high to protect anything. Withstand testing is carried by the insulation system - creepage, potting, the isolation barrier, Y capacitors - while a TVS is a transient energy valve that was never meant to hold off voltage.

Q4: Does repeated dielectric testing damage the TVS?

Yes, and it is a common reason a new board passes while a reworked board does not. Datasheet Ipp is a 10/1000us single-pulse figure, while a 1500 V AC test crosses zero every half cycle - about 3000 avalanche events in one minute - and stretches each event from milliseconds to a minute, a completely different thermal and field-stress regime. The result is IR creeping up, VBR drifting, a higher clamp at high current and eventually a breakdown short. So do not validate parts by repeated voltage application, cap the retests on reworked units, and re-measure IR and VBR at room temperature afterwards.

Contact us

The hard part of a dielectric test is not the numbers, it is the position. The same device is invisible in a differential-mode position and a guaranteed failure in a common-mode position, and both layouts look equally reasonable on a schematic. Two reworks show up most often: blaming the insulation and repeatedly changing creepage, potting and harnesses without fixing anything; and removing the TVS to make the test pass, which does pass the test and removes the protection, until the next ESD or surge event.

Send us your test voltage and pass threshold (DC or AC), the actual test-electrode connections, the current TVS part numbers and mounting positions, and the fixture probe layout, and we will run the five steps in one pass: which devices straddle the withstand path, which can stay, whether to change topology or add a test point, and how to turn the break-and-restore step into a process card. AMSEMI 5.0SMDJ standard and foldback -N datasheets, samples and FAE review are available - please contact us.

Shenzhen Intek Technology Co., Ltd - authorized distributor of AMSEMI TVS and HOPERF CMT digital isolators Tel / WeChat: 136-3264-8484 Address: Room 4 A11, Building 2, No. 309 Ping'an Avenue, Pinghu Street, Longgang District, Shenzhen Website: www.intek.vip

---

Disclaimer: This article is provided for technical education and industry exchange only and does not constitute selection, purchasing or other commercial advice. The 500 V DC, 1000 V DC and 1500 V AC figures and the pass thresholds are examples of common engineering practice; actual values must follow the safety standards, industry standards and company specifications applicable to the product, and this article does not interpret or promise any clause or limit of any standard. The 1500 V AC peak of about 2121 V is an estimate from 1500 x 1.414, and 3000 half cycles in one minute at 50 Hz is a theoretical count; actual test frequency and dwell follow the equipment and the procedure. AMSEMI 5.0SMDJ series Vc and Ipp (datasheets 8060014 standard and 8060055 foldback -N) follow the 10/1000us convention with no 8/20us data; Littelfuse 8x20us figures are that vendor's own convention and the two must not be converted or subtracted. The AMSEMI -N foldback family comes in 3000 W, 5000 W and 8000 W grades that share VBR and Vc at a given voltage and differ only in Ipp; the 8000 W grade is 8.0SMDJ**CA-N (VRWM 48 to 90 V, eleven grades, AEC-Q101, automotive suffix -H), and the voltage ceiling is 100 V for the 5000 W grade and 90 V for the 3000 W and 8000 W grades. Statements about leakage doubling with temperature, TVS failure being predominantly a short, and GDT or MOV behaviour at test voltage are engineering experience and order-of-magnitude descriptions; actual values must follow the manufacturer datasheet, reliability reports and system measurement. High-voltage and battery testing should be performed only by qualified professionals with suitable safety provisions.